Reward based Video Summarization using Advanced Deep Learning Architectures

The goal of video summarization is to produce a short yet precise summary of the original video. Video summary is generated at the end of videos whilst a decision/action needs to be made at every single frame, reinforcement learning is the natural choice for such a job. Even the quality of visual features plays a crucial role in the summary generation, therefore we use advanced deep learning architecture ResNet50 for summarization task. Major contributions in this paper are feature extraction by creating a new dataset and utilizing the newly created dataset for video summarization task using reinforcement learning approach powered by ResNet50 architecture. The experimental results conducted on a benchmark dataset by utilizing a reward-based feedback mechanism achieve the gain of 5.24% for the F1 score in comparison to other state-of-the-art methods in video summarization.